There are more ways to do this than you might guess. The transformation between MNI and Talairach spaces has been done traditionally with what has been called a Brett transform, and that's the one we generally use in our code for converting MNI coordinates to Talairach coordinates, as in the whereami program. The Brett transform is applied in two steps using either two affine transformations or with a couple equations from the original description. More recently, in 2007 and 2010, the Lancaster transform was introduced to accomplish the transformation between these two spaces. This is a single affine transformation, but one has to know which flavor of MNI space is desired: SPM, FSL or other. See these papers for more information:

[dx.doi.org]
[dx.doi.org]
[www3.interscience.wiley.com]
[www3.interscience.wiley.com]

So besides transformations, you also need to be aware of which coordinate order is used. Usually this is either RAI or LPI. "RAI" means incrementing from the right to left, anterior to posterior, inferior to superior. In Talairach and MNI spaces, which are both AC-PC centered, for RAI coordinates, this means the right will be negative, and the left will have positive coordinates. MNI coordinates have generally been reported in LPI coordinates, so that may be some source of confusion. I haven't tried the matlab code you are using, so I can't say anything about the order they assume, but I can suggest some ways to do this using AFNI commands.

First download one of these simple 1D files that describe the various Lancaster transformations from here:

[afni.nimh.nih.gov]
[afni.nimh.nih.gov]

Then using one of these files you can transform the coordinates using these commands as an example:

Vecwarp -input mnicoords.1D -matvec mni2tal.1D > talcoords.1D

or

Vecwarp -input talcoords.1D -matvec mni2tal.1D -backward > mnicoords.1D

Then to put these coordinates as spheres in an AFNI dataset, do something like this:

3dUndump -srad 5 -master ~/abin/TT_N27_EZ_ML+tlrc -xyz \
-prefix tal_sphere_points -orient RAI talcoords.1D

Note that I've used RAI as the coordinate order for both these commands input and output. Viewing the dataset in AFNI, you can easily move among spheres with the Interactive clusterize function in the Overlay panel. Setting the clustering to use Cmass instead of Peak will allow you to go to the center of each sphere quickly. You'll notice the spheres' Cmass centers will always be on a voxel center, and the grid of the dataset may not correspond exactly to the input coordinates.

The Lancaster group also has their GingerAle software along with updated matlab functions to do these same transformations for a list of coordinates.
[www.brainmap.org]
[www.brainmap.org]